[Ffmpeg-devel-irc] ffmpeg-devel.log.20140512

[burek]

[00:01] <smarter> _aca_: you need a patch
[00:02] <smarter> _aca_: http://sprunge.us/VTWO
[00:03] <smarter> (that patch comes from the libav package in Ubuntu originally)
[00:06] <smarter> _aca_: if you want a packaged version you can get mine from https://launchpad.net/~smarter/+archive/ffmpeg but I'm not gonna maintain it
[00:09] <_aca_> michaelni: It builds, but then the symbol ff_resample_int16_rounder is missing from libswresample. Was this intended?
[00:10] <_aca_> smarter: I'm going to try your patch, thanks for the link.
[00:13] <michaelni> no, wasnt intended
[00:14] <michaelni> if smarters solution doesnt work, maybe check what is the difference to existing and working (i assume) DECLARE_ASM_CONST + MANGLE()
[00:15] <cone-266> ffmpeg.git 03Luca Barbato 07master:ebc29519d163: hwaccel: Support specific frame allocators
[00:15] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:c4fe50b954a0: Merge commit 'ebc29519d1634bfeb386c20a5d8a52837aae2436'
[00:29] <cone-266> ffmpeg.git 03Anton Khirnov 07master:dd2d3b766b20: lavc: Add hwaccel private data and init/uninit callbacks
[00:29] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:2ec4586d77c1: Merge commit 'dd2d3b766b20196d0b65a82e3d897ccecbf7adb8'
[00:29] <_aca_> smarter: Thanks! Your patch works great. May I credit you in the changelog?
[00:30] <smarter> _aca_: sure
[00:30] <smarter> maybe say the patch originally comes from the libav packaging too
[00:32] <_aca_> OK.
[00:39] <siretart> _aca_: i guess that's the original patch: http://anonscm.debian.org/gitweb/?p=pkg-multimedia/libav.git;a=commitdiff;h=229f90453d39ec5234a8d7384da4c477c3e01b39
[00:42] <_aca_> siretart: Yes, thanks.
[00:43] <smarter> of course, it'd be better if that patch wasn't needed at all
[00:58] <cone-266> ffmpeg.git 03Anton Khirnov 07master:66e6c8a7b4e3: lavu: Add a pixel format for new VDA hwaccel
[00:58] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:14aef38a6982: Merge commit '66e6c8a7b4e3ac926302529606bd3e323c28b20f'
[01:05] <michaelni> smarter, yes, i also would prefer if we could minimize distro specific changes
[01:06] <smarter> then someone needs to find a way to make ffmpeg build with "LDFLAGS=-Wl,-Bsymbolic-functions ./configure --enable-pic --enable-shared --toolchain=hardened" :]
[01:47] <cone-266> ffmpeg.git 03Anton Khirnov 07master:31a46750c7ea: vda: use hwaccel private data for internal bitstream buffer
[01:47] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:c6a683bf2291: Merge commit '31a46750c7eafe6bce6a3e4f09853cd0a585292b'
[02:32] <cone-266> ffmpeg.git 03Anton Khirnov 07master:67afcefb3593: lavc: Add new VDA hwaccel
[02:32] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:5449239a031e: Merge commit '67afcefb35932b420998f6f3fda46c7c85848a3f'
[02:32] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:2856332719d8: avcodec/vda_h264: fix null pointer dereference
[03:06] <cone-266> ffmpeg.git 03Anton Khirnov 07master:1839fafa9513: avconv: Support VDA hwaccel
[03:06] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:5460ab73d2e8: Merge commit '1839fafa951308237c4672ffd9a94f927b26eac7'
[04:11] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:96abdf2ffd7d: avcodec/vda_h264: fix bistream typo
[04:11] <cone-266> ffmpeg.git 03Matt Oliver 07master:3b2ab8a78306: configure: Allow log2 with msvc12 onwards.
[09:02] <olderz> Ciao a tutti :D
[09:08] <olderz> - sorry is an automatic login message :|
[10:13] <cone-717> ffmpeg.git 03Matt Oliver 07master:4d93e758d732: configure: Ensure log2 is disable based on msvcrt version.
[12:32] <cone-717> ffmpeg.git 03Michael Niedermayer 07master:e951e04a4a69: avcodec/buffersrc: use av_mallocz_array()
[12:32] <cone-717> ffmpeg.git 03Michael Niedermayer 07master:b9cce9ee5c7b: avfilter/buffer: Use av_malloc_array()
[13:43] <plepere> hmmm. I've got a weird problem : I've got a function in ASM. fate gives me OK when compiled with gcc, but a fail when compiled with clang
[13:55] <BBB> plepere: inline asm?
[13:55] <BBB> plepere: show it :)
[13:56] <nevcairiel> is someone fixing windows builds btw? :p
[14:00] <BBB> there's a problem with the windows build?
[14:00] <nevcairiel> well not with the build
[14:00] <BBB> hm, right
[14:00] <nevcairiel> but when you run it and decode hevc, it goes all kaboom
[14:00] <BBB> so mark all functions as taking 16 xmm registers
[14:00] <BBB> not ideal but that's essentially what it does
[14:01] <plepere_> ok, wait a minute BBB
[14:02] <BBB> you can also re-arrange registers
[14:02] <plepere_> it might be a little complicated : it's a big ASM file : http://pastebin.com/wjs4AiAV
[14:03] <plepere_> and hevc_v_loop_filter_luma_10 and hevc_h_loop_filter_luma_10 are the problematic ones
[14:03] <plepere_> but they work in gcc (which is strange because their compiler should be yasm in either way)
[14:06] <BBB> what's the function prototype?
[14:07] <BBB> all your macros already exist in x86util.asm btw, please don't duplicate all these transpose macros
[14:07] <plepere_> ok (it's not my code, BTW)
[14:08] <plepere_> the arguments are (uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q)
[14:08] <BBB> so typically when stuff crashes in a compiler-dependent way, you're either slightly overrunning stack, or slightly overrunning allocated memory
[14:08] <BBB> you could also be clobbering GPRs
[14:08] <plepere_> ok
[14:09] <plepere_> is there an easy way of debugging ?
[14:10] <BBB> you could try valgrind or address sanitizer
[14:10] <BBB> are you using stack?
[14:10] <BBB> seems like you don't?
[14:11] <plepere_> I don't think so, no
[14:12] <BBB> let's first do a bisect
[14:13] <BBB> disable all asm functions in this file except one, and then see if it crashes
[14:13] <BBB> if it does, that's probably your issue
[14:13] <BBB> if not, enable another one
[14:13] <BBB> once you know which entry point causes the crash, let's look at that one closer
[14:14] <plepere_> I've done that : clang gives fate errors when enabling  luma h or v on 10bit
[14:14] <BBB> ok
[14:16] <plepere_> I'll try using valgrind. I've used the tool before
[14:17] <BBB> there's a lot of movdqus in there that should possibly be movdqas
[14:19] <BBB> what is shift_left_param
[14:19] <BBB> does that operate on pointers?
[14:19] <BBB> it seems to
[14:19] <BBB> but it hardcodes their size as 4
[14:19] <BBB> that should be gprsize, not 4
[14:20] <BBB> that would cause memory corruption
[14:20] <BBB> that looks like your bug
[14:20] <BBB>     mov    r7,[%1]
[14:20] <BBB>     mov    r8,[%1+4]
[14:20] <BBB> should be
[14:20] <BBB>     mov    r7,[%1]
[14:20] <BBB>     mov    r8,[%1+gprsize]
[14:20] <BBB> (since gprsize is 8 on x86-64)
[14:20] <BBB> and then same for the store 4 lines down
[14:25] <nevcairiel> that was easy, i found the windows problem
[14:25] <nevcairiel> it punched a gpr in the face!
[14:36] <BBB> that doesn't help indeed :-p
[14:36] <nevcairiel> silly windows, always exposing the developers mistakes
[14:36] <BBB> nevcairiel: patch ok, just commit already
[14:36] <nevcairiel> I cannot
[14:37] <BBB> o
[14:37] <BBB> michaelni: ^^
[14:37] <BBB> need auto-commit robot or so
[14:38] <BBB> if message == 'OK' and person.flags has FLAG.Maintainer: <indent> auto-commit
[14:38] <BBB> was that proper python?
[14:38] <nevcairiel> python needs more whitespace
[14:38] <smarter> "in" instead of "has" too probably
[14:43] <ubitux> BBB: you can remove the indent in a one-liner
[14:44] <ubitux> more like if message == 'OK' and FLAG_MAINTAINER in person.flags: auto_commit() should do it
[14:48] <DHE> Last week I asked for comments on a patch I was making - libx264 to copy closed captions from upstream sources. I've updated it and am seeking new comments: http://www.dehacked.net/ffmpeg-a53-to-libx264.txt
[14:49] <kierank> DHE: ga94 is pretty standard
[14:50] <kierank> DHE: but some of the bits are wrong
[14:50] <kierank> oh you write market bits below
[14:51] <kierank> I would make some of the random numbers #defines
[14:52] <DHE> kierank: in my research (and in the MPEG1/2 decoder) multiple standards are supported. GA94 is just the one I have access to (north american ATSC feeds)
[14:52] <kierank> GA94 is the only standardised one
[14:52] <kierank> the rest are vendor specific
[14:52] <kierank> yes, I have documents about echostar
[14:52] <kierank> and other thingas
[14:52] <kierank> but those in theory are not documents I can share
[14:53] <DHE> I have nothing beyond things I can google, though I have a real IPTV set top box with pretty good access to it
[14:54] <kierank> https://github.com/kierank/broadcastencoder/blob/master/filters/video/cc.c
[14:54] <kierank> that's my implementation
[14:59] <DHE> mpeg1/2 is just reading in the main ccdata payload and saving it, so I'm just writing it back out making up the missing headers as I go. I admit it could be better, but it Works For Me and other people suggested I avoid modifying what the mpeg1/2 decoder collects
[15:07] <kierank> the headers are all documented in atsc and in dvb
[15:07] <kierank> there's no need to make them up
[15:28] <plepere_> BBB : thanks for the help. I'll try that now
[15:28] <plepere_> nevcairiel, thanks for the debugging !
[15:30] <plepere_> BBB : nah, it's not enough. :( I'll keep looking, thanks for the help !
[15:55] <Voicu> I'm sending a stream to a rtmp server and the video packets don't seem to arrive
[15:56] <Voicu> i.e. in VLC I get sound but no video and on the codec information screen there are packets for the audio stream but not for the video one
[15:56] <Voicu> I'm sending video packets with the right stream_index set and the right timestamps
[15:56] <Voicu> (I'm 99% sure about the timestamps)
[15:57] <Voicu> actually nevermind
[17:09] <cone-717> ffmpeg.git 03Hendrik Leppkes 07master:87f2d8079ac6: hevcdsp: correctly indicate that hevc_put_hevc_bi_epel_h uses 9 GPRs
[18:10] <arpu> saste, hello can you help us to port this patches to ffmpeg 2.2 or master http://paste.ubuntu.com/7453051/  (against ffmpeg 1.1) first patchset was from michaelni http://ffmpeg.org/pipermail/ffmpeg-devel/2012-July/127773.html
[18:42] <cone-717> ffmpeg.git 03Carl Eugen Hoyos 07master:a01fedd9b87f: Fix compilation with --enable-lto --enable-shared with icc.
[18:42] <cone-717> ffmpeg.git 03Carl Eugen Hoyos 07master:ab21acecc72a: Use Matroska document version 2 for WebM files if possible.
[18:49] <nevcairiel> ... i fix one issue in hevc, and now some hevc file causes a ICE in msvc 2013u2
[18:50] <Daemon404> laff
[18:51] <nevcairiel> hevcpred_template.c(236) : fatal error C1001: An internal error has occurred in the compiler.
[18:51] <nevcairiel> those are always fun
[18:51] <Daemon404> time to wait N more momnths
[18:51] <Plorkyeran> that's been happening for a while for me
[18:51] <nevcairiel> didnt happen with 2013u1
[18:51] <nevcairiel> i d idnt install any of the pre-releases
[18:52] <Plorkyeran> yeah, ever since updating to 2013u2
[18:52] <nevcairiel> well, the final was just released today
[18:53] <Plorkyeran> ah, I forgot it was a RC
[18:53] <Plorkyeran> they improved linking performance a lot so I upgraded asap
[18:54] <nevcairiel> x64 build seems to work though
[18:54] <Plorkyeran> deleting the 9 and 10 bit versions makes it stop crashing for me
[18:54] <Plorkyeran> I didn't bother investigating further
[18:56] <nevcairiel> maybe the file size hits some magic limit with the inclusion of all these c templates into one compile unit
[18:56] <nevcairiel> lets try using the 64-bit compiler that outputs 32-bit code!
[18:57] <nevcairiel> too bad my fate box doesn't have that one, though
[19:10] <nevcairiel> the code causing the issue is so extremely simple... can'
[19:10] <nevcairiel> can't even change it much
[19:12] <nevcairiel> its a for loop that say "for (i = 0; i < size; i++)" ... if i change that to say "i <=  size-1", it  passes :P
[19:16] <michaelni> saste, is "0510 10:14 Anshul          ( 10K)                       
[19:22] <saste> michaelni, i'll reply on ML in a 20 minutes
[19:40] <Kanerix> Hello
[19:40] <Kanerix> I am having trouble cross-compiling ffmpeg for a specific x86 machine from my x64 machine using a specific toolchain
[19:41] <Kanerix> I think I'm not getting the cross compile paths correct
[19:42] <Kanerix> What argument is --toolchain looking for?
[19:42] <Kanerix> I keep getting unknown toolchain
[19:43] <Kanerix> Is it looking for the binary prefix or the gcc version?
[19:43] <nevcairiel> toolchain is only for a few pre-defined setups
[19:43] <nevcairiel> you can manually pass --cc --ld --as and such
[19:44] <Kanerix> Am I using the full path or relative path to those?
[19:44] <Kanerix> What is --cross-prefix supposed to point to?
[19:45] <nevcairiel> its the name prefix of your cross chain
[19:45] <Kanerix> ah
[19:45] <nevcairiel> ie. i use "x86_64-w64-mingw32-" and gcc is then called x86_64-w64-mingw32-gcc
[19:46] <Kanerix> --sysroot points to which directory?
[19:52] <Kanerix> Trying to shotgun this thing
[20:54] <cone-717> ffmpeg.git 03Michael Niedermayer 07master:d05588e21e47: avformat/utils: limit probe score in the id3 case only if there is a chance that data after the id3 tag will become available
[20:54] <cone-717> ffmpeg.git 03Michael Niedermayer 07master:b3b505c73bf5: avformat/utils: if the id3 tag is too large for data after it to become available, use the full AVPROBE_SCORE_EXTENSION if extension matches
[21:30] <cone-717> ffmpeg.git 03Michael Niedermayer 07master:a8bc175dd9ce: avformat/mp3dec: check the number of frames per buffer size
[22:20] <a141982112> stefano, I have done some changes as you requested, though i am not sure of nit:refactor, can u please check following pastebin
[22:20] <a141982112> http://pastebin.com/CZXVQehf
[22:29] <a141982112> stefano, I am mailing new patch, hope this is what you meant. :)
[23:22] <cone-717> ffmpeg.git 03Carl Eugen Hoyos 07release/2.1:c5a2c26c55eb: Use Matroska document version 2 for WebM files if possible.
[23:22] <cone-717> ffmpeg.git 03Carl Eugen Hoyos 07release/2.2:bb01956d67f9: Use Matroska document version 2 for WebM files if possible.
[23:25] <cone-717> ffmpeg.git 03Michael Niedermayer 07master:81d1fcf37d4b: avcodec: add option to make is_intra_more_likely() from error concealment return "no"
[00:00] --- Tue May 13 2014